Something I forgot to mention is that @ashimgupta95 finetuned hundreds of models and they're available at https://t.co/pmFXO3DLkN
This can be a helpful start in your studies if you don't want to finetune all of these yourself
Explore a jumbo collection of 20+ GPT Prompting Styles for Maximum Impact! 🚀
There's a wealth of tacit knowledge rooted in human intelligence. Thanks to the internet, such wisdom is distilled into comprehensive frameworks that mirror the intricate structures of human creativity
Watch out for our new arxiv preprint where we delve into a question: are the longstanding robustness issues in NLP resolved? I'll give you a glimpse of our findings. 1/
🎉📢 IntenDD, our work on a unified approach towards intent related tasks in conversational systems, leveraging a shared utterance encoding backbone, is now accepted to be part of #EMNLP.
@anas_awadalla Thanks @anas_awadalla, any specific reason why you are using decoder-only models in a generative mode for QA instead of with a span classification head like is typically done? Like here: https://t.co/p8PlKbt394
Of course the causal/symbolic reasoning ability is still missing, and that won't change because of the substrate. But I no longer think it's that important. The vast majority of applications don't need it.
"I went to the referee and said 'look at this - stop the match'."
Sergio Reguilón explains what happened during the incident during the first-half of #NEWTOT...